Description
Quick Overview image1 image2 1SIFT 1SIFT 2matching Image source: YouTube [3:11] 2 Requirement • Implement SIFT and matching using python • Input : given 2 images as a set, 3 sets in total • Output : keypoints of the images and the matching results 3 • Result images should include : • Image pyramid (as shown in page 5) • The DoG of any octave you like (page 7) • The images with keypoints (page 8) • The matching results (page 9) • Write a report : • Explain all parts of your code • Show what you have done in your experiments (e.g., parameters setting) • Result images Quickly go through the requirement Implement SIFT • Construct scale space (image pyramid, difference of gaussian) • Find the local extrema (a.k.a. keypoints) • Generate descriptors 4 Quickly go through the requirement ● Image pyramid for reducing the noise First octave Second octave Third octave Warning: You cannot use cv2.pyrDown() 5 Warning: To contruct image pyramid, you cannot use high level function in any well-developed library. E.g. cv2.pyrDown(), cv2.pyrUp() Quickly go through the requirement ● Image pyramid for reducing the noise : 2″#$ %, ‘%, ‘(%, ⋯ , ‘*#$% , ‘ = 2 -⁄. From Lowe’s paper : number of octaves = 4 number of scale levels = 5 initial σ =1.6 k= 2 as optimal values an octave is actually a set of images were the blur of the last image is double the blur of the first image 1 % ∗ 3 1 ‘% ∗ 3 1 ‘(% ∗ 3 1 ‘4% ∗ 3 1 ‘5% ∗ 3 1 2% ∗ 3 1 2’% ∗ 3 1 2′(% ∗ 3 1 2’4% ∗ 3 1 2’5% ∗ 3 s: # of scale levels in the i th octave 6 ● Difference of Gaussian for reinforcing features Second octave DoG Quickly go through the requirement 7 ● Find the local extrema ● Generate the descriptors Quickly go through the requirement 8 ● Match features between two images For this step, you can use whatever function you want. For example, • RANSAC • Brute-Force Matcher • BLANN Quickly go through the requirement 9 Rules • Implement in python3 • Do not directly call the high level SIFT function in any well-developed library. e.g. cv2.xfeatures2d.SIFT_create() • Do not copy/paste other’s code • Report no more than 10 pages 10 Score • Implement SIFT (45%) • Implement matching (10%) • Report (45%) • We have a scoring example for your matching output (as shown in page 12) • Bonus (9%) • Implement SIFT and matching with your own images • Each set of images would add 3 points at most • 3 sets of images at most 11 Quality example 3 points 2 points 1 points 12 Reminder • You should work on all the given images (3 sets) • Feel free to modify any code provided from TA 13 Submission 14 ● Every submission should consist of the followings: ○ Your code (student_id.py) ○ A readme.txt file describing how to run your code ○ A report (in pdf format) ● Please clip all your files into .zip and submit through New e3